MLflow MCP Server

MLflow MCP Server is an MCP server that connects AI assistants to MLflow tracking servers to manage machine learning workflows. Designed for machine learning engineers, data scientists, and developers using MCP-compliant clients like Claude Desktop or Cursor, this server provides programmatic access to MLflow's experiment tracking and model registry infrastructure. Users can inspect experiments, retrieve individual run parameters and metrics, compare model performance, and track lineage across runs. In addition to experiment tracking, the server enables direct model registry operations, allowing users to browse registered models, create new models or versions from run artifacts, assign deployment aliases such as champion or challenger, and remove outdated models or versions. Operating locally over standard input and output (stdio), the server acts as an operational bridge between LLM interfaces and an active MLflow backend, streamlining model lifecycle management and metric analysis directly through natural language queries.

Category: AI & LLM Tooling

Tags: experiment-tracking, machine learning, mlflow, model-tracking

Visit MLflow MCP Server

How to install and configure MLflow MCP Server

1. Ensure you have Python 3.11+, uv, and an active MLflow tracking server available. 2. Open your client configuration file, such as claude_desktop_config.json for Claude Desktop, ~/.cursor/mcp.json for Cursor, or opencode.json for OpenCode. 3. Add the server entry using uvx under the mcpServers key: json { "mcpServers": { "mlflow": { "command": "uvx", "args": ["mlflow-mcp-server"], "env": { "MLFLOW_TRACKING_URI": "http://localhost:5000" } } } } 4. Update the MLFLOW_TRACKING_URI value if your MLflow server is hosted at a different URL. 5. Save the configuration file and restart your MCP client.

What you can do with MLflow MCP Server

• Comparing metrics, parameters, and tags across multiple experiment runs to evaluate machine learning models. - Finding and filtering experiments by ID or name to locate specific MLflow tracking histories. - Registering new models and creating model versions directly from artifact paths of completed runs. - Managing model deployment stages by assigning aliases like champion or challenger to specific versions. - Deleting deprecated model versions or entire registered models directly through natural language commands.

Key facts

• Open Source
• https://github.com/yesid-lopez/mlflow-mcp-server
• AI & LLM Tooling, Data & Analytics
• experiment-tracking, machine learning, mlflow, model-tracking

Part of MCP Servers

What is MLflow MCP Server?

MLflow MCP Server is a Model Context Protocol tool server that exposes MLflow experiment tracking and model registry functions to AI assistants. It allows language models to fetch metrics, inspect runs, list experiments, and perform model registry updates over a standard input and output interface.

Which MCP clients work with MLflow MCP Server?

MLflow MCP Server works with any standard Model Context Protocol client that supports local command execution over stdio. Documented configurations include Claude Desktop, Cursor, and OpenCode, configured using the uvx runner.

How do I install MLflow MCP Server?

You can run it directly without a manual clone by configuring uvx in your client configuration file, specifying the package mlflow-mcp-server and supplying the MLFLOW_TRACKING_URI environment variable. For local development, clone the repository, run uv sync, and execute the module with uv run -m mlflow_mcp_server.

What tools does MLflow MCP Server provide?

It provides tools for experiment management like get_experiment and search_experiments, run management like get_run and get_experiment_runs, and model registry tools including get_registered_models, create_model_version, set_registered_model_alias, and delete_registered_model.

Is MLflow MCP Server open source?

Yes, MLflow MCP Server is open source and distributed under the MIT license. The source code is publicly accessible on GitHub.